To make a long story short, I ended up simulating an emitter follower with Qucs. This simulator has some transistor models embedded, but you are able to insert PSpice models and subcircuits if you need more.
When you increase input signal level up to the point where the output should clip at zero, one transistor model clips correctly (BFP405, from Qucs) but others not (BFR93a, 2N2222A). The BFR93a model is a PSpice model from the NXP site and I have tried several 2N2222a models (one from Qucs itself, one PSpice model from Zetex and another PSpice model from somewhere): none of these clips at zero although it does at VCC=3.3.
Tried the same circuit in LTSpice with the same result. Am I the only one experiencing this? Has anyone got better large-signal models
The transistor models you can get from NXP are Gummel-Poon models, as are the models that come with LTSpice.
LTspice also supports the superior VBIC - Vertical Bipolar Inter Company - model but manufactures treat the parameters for VBIC models as "commercial in confidence" and don't publish them.
Gummel-Poon doesn't model inverted transistors very accurately, and some time ago I wanted to get hold of a VBIC model to see it it would do better, as it has been claimed that it would, but nobody around here had any non-proprietary data that they were prepared to share with me.
In principle it isn't too difficult to make up a set of VBIC parameters from Gummel-Poon data and a few extra measurements, but nobody seems to have bothered to do it.
Support for GP and VBIC is one thing. You point out how hard it is to get the VBIC model parameters. But even the GP parameters are rarely even close to fully provided.
For example, the models I've seen for BJTs are pretty weak in low current beta. They usually don't include figures for Ise (which defaults to 0) or Ne (defaulted to 1.5, which matters not at all if Ise=0.) [Or in the older literature where these parameters are called C_2=Ise/Is and n_EL.] The models I've seen just assume that a log plot of Ic and Ib versus Vbe is a nice line all the way back towards Vbe=0V, instead of the dramatic knee that actually takes place because actual BJTs have extra effects from recombinations of surface carriers and carriers in the emitter-base space-charge layer, and also due to the formation of emitter-base surface channels. Almost never do I find these low current parameters modeled, nor is there usually enough information in the data sheet to develop them (they start their charts typically with base currents near a microamp, with any serious beta drop-off not shown.) Modelling a discrete equivalent to a PUJT, for example, isn't possible without them because the beta drop at low currents is what makes the difference between a simulation that models real behavior of the discrete pair and one that completely fails and instead finds a stable quiescent point.
I'd just like to see GP parameters, even. Rarely do I see much more than EM2 model parameters, with a few strays from GP added to the mix.
It would be interesting to sweep low base currents from say
5nA to 500nA in order to actually get Ise and Ne to use in simulation. But it's not been a priority of any kind, yet. I just remain aware of the problem. The BJT models that are readily available rarely make use of anything close to all the GP model parameters. And Ise and Ne are merely EM3 model parameters (pre-GP.)
So while it may be not too difficult to add what's needed to get VBIC, assuming you have all the GP parameters, the problem is that you rarely have all the GP parameters. So the difficulty now rises rapidly because of all the holes in the missing modelling parameters for GP or even EM3 in the models that us unwashed mortals can get.
Along with Michael, who has been collecting welfare for longer. He won't admit that it's welfare, any more than I would agree that unemployment benefit was welfare, but it's close enough for this kind of half-baked abuse.
Of course, a professional would have set up a matrix, where you could directly compare the different values assigned to a specific parameter in each different model
Why is that, though? At least, if they ever sell more than
10,000 of them anyway. It takes time, yes. But once for each time they change the process/FAB. How many times is that? Can't be that bad considering all the other time that goes into setting up and manufacturing a line of parts.
Well, they have a personal interest in wanting your work to actually be competitive and work well, too. So of course. Discrete design engineers are mere peons -- and perhaps as a rule not allowed to rely upon a detailed and thorough understanding of the physics involved, so what was learned is often soon forgotten as unused or unreliable instead of being re-inforced every day, as it may otherwise be if a detailed knowledge could be relied upon to make their work more competitive.
I suppose a difference here is that in your work, you are designing for a specific process and FAB situation, so using ALL of the knowledge can help create highly competitive results, whereas people doing discrete design cannot depend on a specific process/FAB being used. So developing specific knowledge doesn't pay off in the same ways. Plus, I suspect that manufacturers would anticipate 100's of times as many phone calls and pestering emails from engineers if they provided the information and they just don't want the hassle, especially since many of the engineers would have forgotten so much and would need "re-education" which itself would cost them still more time and effort. Better to just say less and cut down on pre-sale and after-sale support calls for 5 cent parts, perhaps.
One might argue that designing so that "any old shit" device will work is what real engineering is all about ;-)
In my youth, with help from my oldest son, the programmer, we wrote a bipolar device model parameter extractor.
I suspect it can now be done with Matlab or even Excel... all you need is good data for IC and IB versus VBE and VCE.
I suspect that the OP's problem is just C-B or B-E capacitances that he hadn't considered in his thought processes... "clean" clips only exist in one's imagination >:-} ...Jim Thompson
--
| James E.Thompson, CTO | mens |
| Analog Innovations, Inc. | et |
Not if you want to achieve the best possible results or topologies that depend upon a much fuller knowledge of parameters. It's just not possible for justify how a PUJT arranged discrete BJT pair works, if you don't take into account the reduction in beta as current declines. The topology just doesn't work otherwise. I'm sure there are many other examples. Knowing all the gears within gears, so to speak, allows for a wider range of interesting topologies to unfold, taking advantage of some behavior area or another.
Nifty. I'd love to see it work on the Fairchild datasheet to extract Ne and Ise.
Right. And there is the problem with datasheets. Of course, with a tester, you can get that. But you may need to slew down to 5nA (or less) base currents to extract the slope (Ne) and the intercept (Ise.)
I didn't look at the circuit so I can't add any thoughts.
Our tool was called, appropriately enough, NE.exe ;-)
Datasheets are useless. When I suggested to a major OpAmp manufacturer that the real way to make behavioral models was to have me run the real netlist side-by-side with my behavioral implementation until I attained a complete match... I was basically shown the door :-(
I saved it to my LTspice directory, but haven't tried it yet... too much honey-do in the roll-up to Thanksgiving.
And there's already chit-chat about the Christmas scheduling... we're celebrating it on December 29 this year to accommodate the college students and the granddaughter already graduated and working.
And my sensitive mind suspects something coming I didn't expect to live to see... that _granddaughter_ announcing her engagement ;-)
Maybe I'll make it to great-grandfather :-) ...Jim Thompson
--
| James E.Thompson, CTO | mens |
| Analog Innovations, Inc. | et |
All those models have the same problem w/r to reverse emitter current at 25MHz.
Looks like CJE is four orders of magnitude higher than BFR93a or BFP405.
Correcting this on any of the 2n2222 models fixes it, but I'm not sure how an emitter junction could be expected to exhibit capacitance in the 1E-15 range.
Thanks for the input! At 27 kHz, the 2N2222a clips nicely at the bottom. But, while I expected the 2N2222 to work bad, I would certainly not have expected this for the BFR93a... in an emitter follower!
Otoh, could you expand on the "liberties" that LTspice takes? Sometimes speed is not the most important thing.
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.